Wednesday, October 11, 2017

I ran a number of experiments today, including running a Windows 10 guest with pass-through NVIDIA card and 3D acceleration.

Windows 10 pass-through

When I tried to start my win10-turbo guest, it complained that the PCI cards were already allocated. I had forgotten my attempts at booting non-Linux guests with pass-through.

In the guest configuration, I had simply added:

This is not sufficient to enable the NVIDIA card. I also had to download and install a driver from NVIDIA. After reboot, the primary screen was on the NVIDIA card. Interestingly, I still had QXL as my secondary screen. But there was a catch: when I moved the mouse over the QXL screen, it actually moved the physical cursor over the NVIDIA screen. Which is, actually, a good thing, since it allowed me to run some tests.

Basically, 3D works quite well and smoothly.

Returning that guest to its normal state so that I can resume my streaming tests to investigate the memory leak.

Memory leak

Spent a good fraction of the day building all the GStreamer components in debug mode, using various approaches:

• dnf download followed by mock --rebuild gstreamer1-1.12.2-1.fc26.src.rpm

  * Manually building with rpmbuild * git clone and normal configure and make steps

Finally, as usual, the fastest one (in order to actually debug the problem) is to start with the git repositories. A mock build of a single package takes about 20 minutes.